Merge branch 'main' into adding-support-for-mamba2

ACKNOWLEDGMENTS.md

@@ -14,4 +14,4 @@ MLX Examples was developed with contributions from the following individuals:
 - Markus Enzweiler: Added the `cvae` examples.
 - Prince Canuma: Helped add support for `Starcoder2` models.
 - Shiyu Li: Added the `Segment Anything Model`.
-- Gökdeniz Gülmez: Added support for `MiniCPM`, `Helium`, `Mamba version 1`, `Mamba version 2` and support for `full-fine-tuning`.
+- Gökdeniz Gülmez: Added support for `MiniCPM`, `Helium`, `Mamba version 1`, `Mamba version 2`, `OLMoE` archtectures and support for `full-fine-tuning`.

cifar/README.md

@@ -48,3 +48,17 @@ Note this was run on an M1 Macbook Pro with 16GB RAM.
 
 At the time of writing, `mlx` doesn't have built-in learning rate schedules.
 We intend to update this example once these features are added.
+
+## Distributed training
+
+The example also supports distributed data parallel training. You can launch a
+distributed training as follows:
+
+```shell
+$ cat >hostfile.json
+[
+    {"ssh": "host-to-ssh-to", "ips": ["ip-to-bind-to"]},
+    {"ssh": "host-to-ssh-to", "ips": ["ip-to-bind-to"]}
+]
+$ mlx.launch --verbose --hostfile hostfile.json main.py --batch 256 --epochs 5 --arch resnet20
+```

cifar/dataset.py

@@ -1,3 +1,4 @@
+import mlx.core as mx
 import numpy as np
 from mlx.data.datasets import load_cifar10
 
@@ -12,8 +13,11 @@ def get_cifar10(batch_size, root=None):
         x = x.astype("float32") / 255.0
         return (x - mean) / std
 
+    group = mx.distributed.init()
+
     tr_iter = (
         tr.shuffle()
+        .partition_if(group.size() > 1, group.size(), group.rank())
         .to_stream()
         .image_random_h_flip("image", prob=0.5)
         .pad("image", 0, 4, 4, 0.0)
@@ -25,6 +29,11 @@ def get_cifar10(batch_size, root=None):
     )
 
     test = load_cifar10(root=root, train=False)
-    test_iter = test.to_stream().key_transform("image", normalize).batch(batch_size)
+    test_iter = (
+        test.to_stream()
+        .partition_if(group.size() > 1, group.size(), group.rank())
+        .key_transform("image", normalize)
+        .batch(batch_size)
+    )
 
     return tr_iter, test_iter

cifar/main.py

@@ -23,6 +23,13 @@ parser.add_argument("--seed", type=int, default=0, help="random seed")
 parser.add_argument("--cpu", action="store_true", help="use cpu only")
 
 
+def print_zero(group, *args, **kwargs):
+    if group.rank() != 0:
+        return
+    flush = kwargs.pop("flush", True)
+    print(*args, **kwargs, flush=flush)
+
+
 def eval_fn(model, inp, tgt):
     return mx.mean(mx.argmax(model(inp), axis=1) == tgt)
 
@@ -34,9 +41,20 @@ def train_epoch(model, train_iter, optimizer, epoch):
         acc = mx.mean(mx.argmax(output, axis=1) == tgt)
         return loss, acc
 
-    losses = []
+    world = mx.distributed.init()
-    accs = []
+    losses = 0
-    samples_per_sec = []
+    accuracies = 0
+    samples_per_sec = 0
+    count = 0
+
+    def average_stats(stats, count):
+        if world.size() == 1:
+            return [s / count for s in stats]
+
+        with mx.stream(mx.cpu):
+            stats = mx.distributed.all_sum(mx.array(stats))
+            count = mx.distributed.all_sum(count)
+            return (stats / count).tolist()
 
     state = [model.state, optimizer.state]
 
@@ -44,6 +62,7 @@ def train_epoch(model, train_iter, optimizer, epoch):
     def step(inp, tgt):
         train_step_fn = nn.value_and_grad(model, train_step)
         (loss, acc), grads = train_step_fn(model, inp, tgt)
+        grads = nn.utils.average_gradients(grads)
         optimizer.update(model, grads)
         return loss, acc
 
@@ -52,69 +71,79 @@ def train_epoch(model, train_iter, optimizer, epoch):
         y = mx.array(batch["label"])
         tic = time.perf_counter()
         loss, acc = step(x, y)
-        mx.eval(state)
+        mx.eval(loss, acc, state)
         toc = time.perf_counter()
-        loss = loss.item()
+        losses += loss.item()
-        acc = acc.item()
+        accuracies += acc.item()
-        losses.append(loss)
+        samples_per_sec += x.shape[0] / (toc - tic)
-        accs.append(acc)
+        count += 1
-        throughput = x.shape[0] / (toc - tic)
-        samples_per_sec.append(throughput)
         if batch_counter % 10 == 0:
-            print(
+            l, a, s = average_stats(
+                [losses, accuracies, world.size() * samples_per_sec],
+                count,
+            )
+            print_zero(
+                world,
                 " | ".join(
                     (
                         f"Epoch {epoch:02d} [{batch_counter:03d}]",
-                        f"Train loss {loss:.3f}",
+                        f"Train loss {l:.3f}",
-                        f"Train acc {acc:.3f}",
+                        f"Train acc {a:.3f}",
-                        f"Throughput: {throughput:.2f} images/second",
+                        f"Throughput: {s:.2f} images/second",
-                    )
                     )
+                ),
             )
 
-    mean_tr_loss = mx.mean(mx.array(losses))
+    return average_stats([losses, accuracies, world.size() * samples_per_sec], count)
-    mean_tr_acc = mx.mean(mx.array(accs))
-    samples_per_sec = mx.mean(mx.array(samples_per_sec))
-    return mean_tr_loss, mean_tr_acc, samples_per_sec
 
 
 def test_epoch(model, test_iter, epoch):
-    accs = []
+    accuracies = 0
+    count = 0
     for batch_counter, batch in enumerate(test_iter):
         x = mx.array(batch["image"])
         y = mx.array(batch["label"])
         acc = eval_fn(model, x, y)
-        acc_value = acc.item()
+        accuracies += acc.item()
-        accs.append(acc_value)
+        count += 1
-    mean_acc = mx.mean(mx.array(accs))
+
-    return mean_acc
+    with mx.stream(mx.cpu):
+        accuracies = mx.distributed.all_sum(accuracies)
+        count = mx.distributed.all_sum(count)
+        return (accuracies / count).item()
 
 
 def main(args):
     mx.random.seed(args.seed)
 
+    # Initialize the distributed group and report the nodes that showed up
+    world = mx.distributed.init()
+    if world.size() > 1:
+        print(f"Starting rank {world.rank()} of {world.size()}", flush=True)
+
     model = getattr(resnet, args.arch)()
 
-    print("Number of params: {:0.04f} M".format(model.num_params() / 1e6))
+    print_zero(world, f"Number of params: {model.num_params() / 1e6:0.04f} M")
 
     optimizer = optim.Adam(learning_rate=args.lr)
 
     train_data, test_data = get_cifar10(args.batch_size)
     for epoch in range(args.epochs):
         tr_loss, tr_acc, throughput = train_epoch(model, train_data, optimizer, epoch)
-        print(
+        print_zero(
+            world,
             " | ".join(
                 (
                     f"Epoch: {epoch}",
-                    f"avg. Train loss {tr_loss.item():.3f}",
+                    f"avg. Train loss {tr_loss:.3f}",
-                    f"avg. Train acc {tr_acc.item():.3f}",
+                    f"avg. Train acc {tr_acc:.3f}",
-                    f"Throughput: {throughput.item():.2f} images/sec",
+                    f"Throughput: {throughput:.2f} images/sec",
-                )
                 )
+            ),
         )
 
         test_acc = test_epoch(model, test_data, epoch)
-        print(f"Epoch: {epoch} | Test acc {test_acc.item():.3f}")
+        print_zero(world, f"Epoch: {epoch} | Test acc {test_acc:.3f}")
 
         train_data.reset()
         test_data.reset()

llms/mlx_lm/_version.py

@@ -1,3 +1,3 @@
 # Copyright © 2023-2024 Apple Inc.
 
-__version__ = "0.21.5"
+__version__ = "0.21.6"

llms/mlx_lm/chat.py

@@ -11,7 +11,7 @@ from .utils import load, stream_generate
 
 DEFAULT_TEMP = 0.0
 DEFAULT_TOP_P = 1.0
-DEFAULT_SEED = 0
+DEFAULT_SEED = None
 DEFAULT_MAX_TOKENS = 256
 DEFAULT_MODEL = "mlx-community/Llama-3.2-3B-Instruct-4bit"
 
@@ -36,7 +36,12 @@ def setup_arg_parser():
     parser.add_argument(
         "--top-p", type=float, default=DEFAULT_TOP_P, help="Sampling top-p"
     )
-    parser.add_argument("--seed", type=int, default=DEFAULT_SEED, help="PRNG seed")
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=DEFAULT_SEED,
+        help="PRNG seed",
+    )
     parser.add_argument(
         "--max-kv-size",
         type=int,
@@ -57,6 +62,7 @@ def main():
     parser = setup_arg_parser()
     args = parser.parse_args()
 
+    if args.seed is not None:
         mx.random.seed(args.seed)
 
     model, tokenizer = load(
@@ -65,12 +71,25 @@ def main():
         tokenizer_config={"trust_remote_code": True},
     )
 
-    print(f"[INFO] Starting chat session with {args.model}. To exit, enter 'q'.")
+    def print_help():
+        print("The command list:")
+        print("- 'q' to exit")
+        print("- 'r' to reset the chat")
+        print("- 'h' to display these commands")
+
+    print(f"[INFO] Starting chat session with {args.model}.")
+    print_help()
     prompt_cache = make_prompt_cache(model, args.max_kv_size)
     while True:
         query = input(">> ")
         if query == "q":
             break
+        if query == "r":
+            prompt_cache = make_prompt_cache(model, args.max_kv_size)
+            continue
+        if query == "h":
+            print_help()
+            continue
         messages = [{"role": "user", "content": query}]
         prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
         for response in stream_generate(

llms/mlx_lm/convert.py

@@ -2,8 +2,23 @@
 
 import argparse
 
+from . import utils
 from .utils import convert
 
+QUANT_RECIPES = [
+    "mixed_2_6",
+    "mixed_3_6",
+]
+
+
+def quant_args(arg):
+    if arg not in QUANT_RECIPES:
+        raise argparse.ArgumentTypeError(
+            f"Invalid q-recipe {arg!r}. Choose from: {QUANT_RECIPES}"
+        )
+    else:
+        return getattr(utils, arg)
+
 
 def configure_parser() -> argparse.ArgumentParser:
     """
@@ -29,6 +44,12 @@ def configure_parser() -> argparse.ArgumentParser:
     parser.add_argument(
         "--q-bits", help="Bits per weight for quantization.", type=int, default=4
     )
+    parser.add_argument(
+        "--quant-predicate",
+        help=f"Mixed-bit quantization recipe. Choices: {QUANT_RECIPES}",
+        type=quant_args,
+        required=False,
+    )
     parser.add_argument(
         "--dtype",
         help="Type to save the non-quantized parameters.",

llms/mlx_lm/evaluate.py

@@ -289,17 +289,15 @@ class MLXLM(LM):
         contexts, options = zip(*[req.args for req in requests])
         # contrary to the doc the second element of the tuple contains
         # {'do_sample': False, 'until': ['\n\n'], 'temperature': 0}
-        keys = list(options[0].keys())
-        assert "until" in keys
-        untils = [x["until"] for x in options]
         completions = []
 
-        for context, until in tqdm(zip(contexts, untils), total=len(contexts)):
+        for context, opt in tqdm(zip(contexts, options), total=len(contexts)):
+            until = opt["until"]
             context = self.tokenizer.encode(
                 context, add_special_tokens=not self.use_chat_template
             )
             max_tokens = min(
-                self._max_tokens,
+                opt.get("max_gen_tokens", self._max_tokens),
                 self.tokenizer.model_max_length - len(context),
             )
             text = ""
@@ -334,9 +332,9 @@ def main():
     )
     parser.add_argument(
         "--limit",
-        default=1.0,
+        default=100,
         help="Limit the number of examples per task.",
-        type=float,
+        type=int,
     )
     parser.add_argument("--seed", type=int, default=123, help="Random seed.")
     parser.add_argument(

llms/mlx_lm/examples/lora_config.yaml

@@ -7,6 +7,15 @@ train: true
 # The fine-tuning method: "lora", "dora", or "full".
 fine_tune_type: lora
 
+# The Optimizer with its possible inputs
+optimizer: adamw
+# optimizer_config:
+#   adamw:
+#     betas: [0.9, 0.98]
+#     eps: 1e-6
+#     weight_decay: 0.05
+#     bias_correction: true
+
 # Directory with {train, valid, test}.jsonl files
 data: "/path/to/training/data"
 

llms/mlx_lm/examples/tool_use.py

@@ -0,0 +1,73 @@
+# Copyright © 2025 Apple Inc.
+
+import json
+
+from mlx_lm import generate, load
+from mlx_lm.models.cache import make_prompt_cache
+
+# Specify the checkpoint
+checkpoint = "mlx-community/Qwen2.5-32B-Instruct-4bit"
+
+# Load the corresponding model and tokenizer
+model, tokenizer = load(path_or_hf_repo=checkpoint)
+
+
+# An example tool, make sure to include a docstring and type hints
+def multiply(a: float, b: float):
+    """
+    A function that multiplies two numbers
+
+    Args:
+        a: The first number to multiply
+        b: The second number to multiply
+    """
+    return a * b
+
+
+tools = {"multiply": multiply}
+
+# Specify the prompt and conversation history
+prompt = "Multiply 12234585 and 48838483920."
+messages = [{"role": "user", "content": prompt}]
+
+prompt = tokenizer.apply_chat_template(
+    messages, add_generation_prompt=True, tools=list(tools.values())
+)
+
+prompt_cache = make_prompt_cache(model)
+
+# Generate the initial tool call:
+response = generate(
+    model=model,
+    tokenizer=tokenizer,
+    prompt=prompt,
+    max_tokens=2048,
+    verbose=True,
+    prompt_cache=prompt_cache,
+)
+
+# Parse the tool call:
+# (Note, the tool call format is model specific)
+tool_open = "<tool_call>"
+tool_close = "</tool_call>"
+start_tool = response.find(tool_open) + len(tool_open)
+end_tool = response.find(tool_close)
+tool_call = json.loads(response[start_tool:end_tool].strip())
+tool_result = tools[tool_call["name"]](**tool_call["arguments"])
+
+# Put the tool result in the prompt
+messages = [{"role": "tool", "name": tool_call["name"], "content": tool_result}]
+prompt = tokenizer.apply_chat_template(
+    messages,
+    add_generation_prompt=True,
+)
+
+# Generate the final response:
+response = generate(
+    model=model,
+    tokenizer=tokenizer,
+    prompt=prompt,
+    max_tokens=2048,
+    verbose=True,
+    prompt_cache=prompt_cache,
+)

llms/mlx_lm/generate.py

@@ -16,7 +16,7 @@ DEFAULT_TEMP = 0.0
 DEFAULT_TOP_P = 1.0
 DEFAULT_MIN_P = 0.0
 DEFAULT_MIN_TOKENS_TO_KEEP = 1
-DEFAULT_SEED = 0
+DEFAULT_SEED = None
 DEFAULT_MODEL = "mlx-community/Llama-3.2-3B-Instruct-4bit"
 DEFAULT_QUANTIZED_KV_START = 5000
 
@@ -60,6 +60,11 @@ def setup_arg_parser():
         default=DEFAULT_PROMPT,
         help="Message to be processed by the model ('-' reads from stdin)",
     )
+    parser.add_argument(
+        "--prefill-response",
+        default=None,
+        help="Prefill response to be used for the chat template",
+    )
     parser.add_argument(
         "--max-tokens",
         "-m",
@@ -82,7 +87,12 @@ def setup_arg_parser():
         default=DEFAULT_MIN_TOKENS_TO_KEEP,
         help="Minimum tokens to keep for min-p sampling.",
     )
-    parser.add_argument("--seed", type=int, default=DEFAULT_SEED, help="PRNG seed")
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=DEFAULT_SEED,
+        help="PRNG seed",
+    )
     parser.add_argument(
         "--ignore-chat-template",
         action="store_true",
@@ -147,7 +157,7 @@ def setup_arg_parser():
         "--num-draft-tokens",
         type=int,
         help="Number of tokens to draft when using speculative decoding.",
-        default=2,
+        default=3,
     )
     return parser
 
@@ -155,6 +165,8 @@ def setup_arg_parser():
 def main():
     parser = setup_arg_parser()
     args = parser.parse_args()
+
+    if args.seed is not None:
         mx.random.seed(args.seed)
 
     # Load the prompt cache and metadata if a cache file is provided
@@ -219,10 +231,14 @@ def main():
             messages = []
         messages.append({"role": "user", "content": prompt})
 
+        has_prefill = args.prefill_response is not None
+        if has_prefill:
+            messages.append({"role": "assistant", "content": args.prefill_response})
         prompt = tokenizer.apply_chat_template(
             messages,
             tokenize=False,
-            add_generation_prompt=True,
+            continue_final_message=has_prefill,
+            add_generation_prompt=not has_prefill,
             **template_kwargs,
         )
 
@@ -233,7 +249,8 @@ def main():
             test_prompt = tokenizer.apply_chat_template(
                 messages,
                 tokenize=False,
-                add_generation_prompt=True,
+                continue_final_message=has_prefill,
+                add_generation_prompt=not has_prefill,
             )
             prompt = prompt[test_prompt.index("<query>") :]
         prompt = tokenizer.encode(prompt, add_special_tokens=False)

llms/mlx_lm/lora.py

@@ -43,6 +43,11 @@ CONFIG_DEFAULTS = {
     "model": "mlx_model",
     "train": False,
     "fine_tune_type": "lora",
+    "optimizer": "adam",
+    "optimizer_config": {
+        "adam": {},
+        "adamw": {},
+    },
     "data": "data/",
     "seed": 0,
     "num_layers": 16,
@@ -62,6 +67,7 @@ CONFIG_DEFAULTS = {
     "grad_checkpoint": False,
     "lr_schedule": None,
     "lora_parameters": {"rank": 8, "alpha": 16, "dropout": 0.0, "scale": 10.0},
+    "mask_prompt": False,
 }
 
 
@@ -94,14 +100,19 @@ def build_parser():
         choices=["lora", "dora", "full"],
         help="Type of fine-tuning to perform: lora, dora, or full.",
     )
-
+    parser.add_argument(
+        "--optimizer",
+        type=str,
+        choices=["adam", "adamw"],
+        default=None,
+        help="Optimizer to use for training: adam or adamw",
+    )
     parser.add_argument(
         "--mask-prompt",
         action="store_true",
         help="Mask the prompt in the loss when training",
-        default=False,
+        default=None,
     )
-
     parser.add_argument(
         "--num-layers",
         type=int,
@@ -228,11 +239,21 @@ def train_model(
     )
 
     model.train()
-    opt = optim.Adam(
+
-        learning_rate=(
+    # Initialize the selected optimizer
-            build_schedule(args.lr_schedule) if args.lr_schedule else args.learning_rate
+    lr = build_schedule(args.lr_schedule) if args.lr_schedule else args.learning_rate
-        )
+
-    )
+    optimizer_name = args.optimizer.lower()
+    optimizer_config = args.optimizer_config.get(optimizer_name, {})
+
+    if optimizer_name == "adam":
+        opt_class = optim.Adam
+    elif optimizer_name == "adamw":
+        opt_class = optim.AdamW
+    else:
+        raise ValueError(f"Unsupported optimizer: {optimizer_name}")
+
+    opt = opt_class(learning_rate=lr, **optimizer_config)
 
     # Train model
     train(

llms/mlx_lm/manage.py

@@ -2,7 +2,22 @@ import argparse
 from typing import List, Union
 
 from huggingface_hub import scan_cache_dir
-from transformers.commands.user import tabulate
+
+
+def tabulate(rows: List[List[Union[str, int]]], headers: List[str]) -> str:
+    """
+    Inspired by:
+    - stackoverflow.com/a/8356620/593036
+    - stackoverflow.com/questions/9535954/printing-lists-as-tabular-data
+    """
+    col_widths = [max(len(str(x)) for x in col) for col in zip(*rows, headers)]
+    row_format = ("{{:{}}} " * len(headers)).format(*col_widths)
+    lines = []
+    lines.append(row_format.format(*headers))
+    lines.append(row_format.format(*["-" * w for w in col_widths]))
+    for row in rows:
+        lines.append(row_format.format(*row))
+    return "\n".join(lines)
 
 
 def ask_for_confirmation(message: str) -> bool:

llms/mlx_lm/models/base.py

@@ -33,13 +33,13 @@ def create_causal_mask(
     linds = mx.arange(offset, offset + N) if offset else rinds
     linds = linds[:, None]
     rinds = rinds[None]
-    mask = linds < rinds
+    mask = linds >= rinds
     if window_size is not None:
-        mask = mask | (linds > rinds + window_size)
+        mask = mask & (linds <= rinds + window_size)
     if lengths is not None:
         lengths = lengths[:, None, None, None]
-        mask = mask | (rinds >= lengths)
+        mask = mask & (rinds < lengths)
-    return mask * -1e9
+    return mask
 
 
 def create_attention_mask(h: mx.array, cache: Optional[Any] = None):
@@ -55,7 +55,6 @@ def create_attention_mask(h: mx.array, cache: Optional[Any] = None):
             else:
                 offset = c.offset
         mask = create_causal_mask(T, offset, window_size=window_size)
-        mask = mask.astype(h.dtype)
     else:
         mask = None
     return mask

llms/mlx_lm/models/deepseek_v3.py

@@ -181,6 +181,7 @@ class DeepseekV3Attention(nn.Module):
             bias=config.attention_bias,
         )
 
+        if self.config.rope_scaling is not None:
             mscale_all_dim = self.config.rope_scaling.get("mscale_all_dim", 0)
             scaling_factor = self.config.rope_scaling["factor"]
             if mscale_all_dim:
@@ -205,6 +206,12 @@ class DeepseekV3Attention(nn.Module):
                 base=self.rope_theta,
                 **rope_kwargs,
             )
+        else:
+            self.rope = nn.RoPE(
+                dims=self.qk_rope_head_dim,
+                base=self.rope_theta,
+                traditional=True,
+            )
 
     def __call__(
         self,
@@ -487,8 +494,12 @@ class Model(nn.Module):
                         ]
                         weights[f"{prefix}.mlp.switch_mlp.{m}.{k}"] = mx.stack(to_join)
 
-        # Remove multi-token prediction layer
+        # Remove multi-token prediction layer and any unused precomputed rotary freqs
-        return {k: v for k, v in weights.items() if not k.startswith("model.layers.61")}
+        return {
+            k: v
+            for k, v in weights.items()
+            if not k.startswith("model.layers.61") and "rotary_emb.inv_freq" not in k
+        }
 
     @property
     def layers(self):

llms/mlx_lm/models/llama.py

@@ -196,9 +196,12 @@ class Model(nn.Module):
 
     def sanitize(self, weights):
         # Remove unused precomputed rotary freqs
-        return {
+        weights = {
             k: v for k, v in weights.items() if "self_attn.rotary_emb.inv_freq" not in k
         }
+        if self.args.tie_word_embeddings:
+            weights.pop("lm_head.weight", None)
+        return weights
 
     @property
     def layers(self):

llms/mlx_lm/models/olmoe.py

@@ -0,0 +1,217 @@
+# Copyright © 2023-2024 Apple Inc.
+
+from dataclasses import dataclass
+from typing import Any, Dict, Optional, Union
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .base import BaseModelArgs, create_attention_mask, scaled_dot_product_attention
+from .rope_utils import initialize_rope
+from .switch_layers import SwitchGLU
+
+
+@dataclass
+class ModelArgs(BaseModelArgs):
+    model_type: str
+    hidden_size: int
+    num_hidden_layers: int
+    intermediate_size: int
+    num_attention_heads: int
+    rms_norm_eps: float
+    vocab_size: int
+    num_experts: int
+    num_experts_per_tok: int
+    norm_topk_prob: bool = False
+    head_dim: Optional[int] = None
+    max_position_embeddings: Optional[int] = None
+    num_key_value_heads: Optional[int] = None
+    attention_bias: bool = False
+    mlp_bias: bool = False
+    rope_theta: float = 10000
+    rope_traditional: bool = False
+    rope_scaling: Optional[Dict[str, Union[float, str]]] = None
+    tie_word_embeddings: bool = True
+
+    def __post_init__(self):
+        if self.num_key_value_heads is None:
+            self.num_key_value_heads = self.num_attention_heads
+
+
+class Attention(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+
+        dim = args.hidden_size
+        self.n_heads = n_heads = args.num_attention_heads
+        self.n_kv_heads = n_kv_heads = args.num_key_value_heads
+
+        self.head_dim = head_dim = args.head_dim or args.hidden_size // n_heads
+
+        self.scale = head_dim**-0.5
+
+        self.q_proj = nn.Linear(dim, n_heads * head_dim, bias=args.attention_bias)
+        self.k_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=args.attention_bias)
+        self.v_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=args.attention_bias)
+        self.o_proj = nn.Linear(n_heads * head_dim, dim, bias=args.attention_bias)
+
+        self.rope = initialize_rope(
+            self.head_dim,
+            args.rope_theta,
+            args.rope_traditional,
+            args.rope_scaling,
+            args.max_position_embeddings,
+        )
+
+        self.q_norm = nn.RMSNorm(n_heads * head_dim, args.rms_norm_eps)
+        self.k_norm = nn.RMSNorm(n_kv_heads * head_dim, args.rms_norm_eps)
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array:
+        B, L, D = x.shape
+        queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)
+        queries = self.q_norm(queries)
+        keys = self.k_norm(keys)
+        queries = queries.reshape(B, L, self.n_heads, -1).transpose(0, 2, 1, 3)
+        keys = keys.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
+        values = values.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
+        if cache is not None:
+            queries = self.rope(queries, offset=cache.offset)
+            keys = self.rope(keys, offset=cache.offset)
+            keys, values = cache.update_and_fetch(keys, values)
+        else:
+            queries = self.rope(queries)
+            keys = self.rope(keys)
+        output = scaled_dot_product_attention(
+            queries, keys, values, cache=cache, scale=self.scale, mask=mask
+        )
+        output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
+        return self.o_proj(output)
+
+
+class OlmoeSparseMoeBlock(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.num_experts = args.num_experts
+        self.top_k = args.num_experts_per_tok
+        self.norm_topk_prob = args.norm_topk_prob
+
+        self.gate = nn.Linear(args.hidden_size, self.num_experts, bias=False)
+        self.switch_mlp = SwitchGLU(
+            args.hidden_size,
+            args.intermediate_size,
+            self.num_experts,
+            bias=args.mlp_bias,
+        )
+
+    def __call__(self, x: mx.array) -> mx.array:
+        B, L, D = x.shape
+        x_flat = x.reshape(-1, D)
+        router_logits = self.gate(x_flat)
+        routing_weights = mx.softmax(router_logits, axis=1, precise=True)
+        k = self.top_k
+        indices = mx.stop_gradient(
+            mx.argpartition(-routing_weights, kth=k - 1, axis=-1)[..., :k]
+        )
+        scores = mx.take_along_axis(routing_weights, indices, axis=-1)
+        if self.norm_topk_prob:
+            scores = scores / scores.sum(axis=-1, keepdims=True)
+        y = self.switch_mlp(x_flat, indices)
+        y = (y * scores[..., None]).sum(axis=-2)
+        return y.reshape(B, L, D)
+
+
+class TransformerBlock(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.self_attn = Attention(args)
+        self.mlp = OlmoeSparseMoeBlock(args)
+        self.input_layernorm = nn.RMSNorm(args.hidden_size, eps=args.rms_norm_eps)
+        self.post_attention_layernorm = nn.RMSNorm(
+            args.hidden_size, eps=args.rms_norm_eps
+        )
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array:
+        x = x + self.self_attn(self.input_layernorm(x), mask, cache)
+        x = x + self.mlp(self.post_attention_layernorm(x))
+        return x
+
+
+class OlmoeModel(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.args = args
+        self.vocab_size = args.vocab_size
+        self.num_hidden_layers = args.num_hidden_layers
+        assert self.vocab_size > 0
+        self.embed_tokens = nn.Embedding(args.vocab_size, args.hidden_size)
+        self.layers = [
+            TransformerBlock(args=args) for _ in range(args.num_hidden_layers)
+        ]
+        self.norm = nn.RMSNorm(args.hidden_size, eps=args.rms_norm_eps)
+
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache=None,
+        mask=None,
+    ):
+        h = self.embed_tokens(inputs)
+        if mask is None:
+            mask = create_attention_mask(h, cache)
+        if cache is None:
+            cache = [None] * len(self.layers)
+        for layer, c in zip(self.layers, cache):
+            h = layer(h, mask, cache=c)
+        return self.norm(h)
+
+
+class Model(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.args = args
+        self.model_type = args.model_type
+        self.model = OlmoeModel(args)
+        if not args.tie_word_embeddings:
+            self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)
+
+    def __call__(
+        self,
+        inputs: mx.array,
+        cache=None,
+        mask=None,
+    ):
+        out = self.model(inputs, cache, mask)
+        if self.args.tie_word_embeddings:
+            out = self.model.embed_tokens.as_linear(out)
+        else:
+            out = self.lm_head(out)
+        return out
+
+    def sanitize(self, weights):
+        if "model.layers.0.mlp.experts.0.up_proj.weight" not in weights:
+            return weights
+        for l in range(self.args.num_hidden_layers):
+            prefix = f"model.layers.{l}"
+            for n in ["up_proj", "down_proj", "gate_proj"]:
+                for k in ["weight", "scales", "biases"]:
+                    if f"{prefix}.mlp.experts.0.{n}.{k}" in weights:
+                        to_join = [
+                            weights.pop(f"{prefix}.mlp.experts.{e}.{n}.{k}")
+                            for e in range(self.args.num_experts)
+                        ]
+                        weights[f"{prefix}.mlp.switch_mlp.{n}.{k}"] = mx.stack(to_join)
+        return weights
+
+    @property
+    def layers(self):
+        return self.model.layers

llms/mlx_lm/models/phi3.py

@@ -23,8 +23,10 @@ class ModelArgs(BaseModelArgs):
     rope_theta: float = 10000
     rope_traditional: bool = False
     rope_scaling: Optional[Dict[str, Union[float, List[float]]]] = None
+    partial_rotary_factor: float = 1.0
     max_position_embeddings: int = 131072
     original_max_position_embeddings: int = 4096
+    tie_word_embeddings: bool = False
 
     def __post_init__(self):
         if self.num_key_value_heads is None:
@@ -59,9 +61,10 @@ class Attention(nn.Module):
         self.qkv_proj = nn.Linear(dim, op_size, bias=False)
         self.o_proj = nn.Linear(n_heads * head_dim, dim, bias=False)
 
+        rope_dim = int(head_dim * args.partial_rotary_factor)
         if args.rope_scaling and args.rope_scaling["type"] in ["longrope", "su"]:
             self.rope = SuScaledRotaryEmbedding(
-                head_dim,
+                rope_dim,
                 base=args.rope_theta,
                 max_position_embeddings=args.max_position_embeddings,
                 original_max_position_embeddings=args.original_max_position_embeddings,
@@ -74,7 +77,7 @@ class Attention(nn.Module):
                 assert isinstance(args.rope_scaling["factor"], float)
                 rope_scale = 1 / args.rope_scaling["factor"]
             self.rope = nn.RoPE(
-                head_dim,
+                rope_dim,
                 traditional=args.rope_traditional,
                 base=args.rope_theta,
                 scale=rope_scale,
@@ -190,6 +193,7 @@ class Model(nn.Module):
         super().__init__()
         self.model_type = args.model_type
         self.model = Phi3Model(args)
+        if not args.tie_word_embeddings:
             self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)
         self.args = args
 
@@ -200,7 +204,11 @@ class Model(nn.Module):
         cache=None,
     ):
         out = self.model(inputs, mask, cache)
-        return self.lm_head(out)
+        if self.args.tie_word_embeddings:
+            out = self.model.embed_tokens.as_linear(out)
+        else:
+            out = self.lm_head(out)
+        return out
 
     @property
     def layers(self):

llms/mlx_lm/models/plamo2.py

@@ -2,7 +2,7 @@
 
 import math
 from dataclasses import dataclass
-from typing import Any, Optional, Union
+from typing import Any, Optional
 
 import mlx.core as mx
 import mlx.nn as nn
@@ -32,7 +32,6 @@ class ModelArgs(BaseModelArgs):
     mamba_enabled: bool = True
     intermediate_size: int = 13312
     vocab_size: int = 32000
-    max_position_embeddings: int = 10 * 1024 * 1024
 
 
 class RMSNorm(nn.Module):
@@ -53,6 +52,16 @@ class RMSNorm(nn.Module):
         )
 
 
+def _rms_norm(hidden_states: mx.array, eps: float) -> mx.array:
+    input_dtype = hidden_states.dtype
+    hidden_states = hidden_states.astype(mx.float32)
+    variance = mx.power(hidden_states, 2).mean(-1, keepdims=True)
+    hidden_states = hidden_states * mx.rsqrt(variance + eps)
+    hidden_states = hidden_states.astype(input_dtype)
+
+    return hidden_states
+
+
 def get_initial_dt_bias(num_heads: int) -> mx.array:
     dt_min = 0.001
     dt_max = 0.1
@@ -220,8 +229,7 @@ def ssd_chunk_scan_combined(
 
 
 def causal_conv1d_update(conv_state, x, weight) -> tuple[mx.array, mx.array]:
-    batch, seqlen, dim = x.shape
+    _, seqlen, dim = x.shape
-    width = weight.shape[1]
     state_len = conv_state.shape[-2]
     x = mx.concatenate([conv_state, x], axis=-2)
     conv_state = x[:, -state_len:]
@@ -392,8 +400,8 @@ class Attention(nn.Module):
         k = k.reshape(B, T, self.k_num_heads, self.qk_dim).transpose(0, 2, 1, 3)
         v = v.reshape(B, T, self.v_num_heads, self.v_dim).transpose(0, 2, 1, 3)
 
-        q = mx.fast.layer_norm(q, None, None, 1e-6) * self.q_weight[:, None]
+        q = _rms_norm(q, 1e-6) * self.q_weight[:, None]
-        k = mx.fast.layer_norm(k, None, None, 1e-6) * self.k_weight[:, None]
+        k = _rms_norm(k, 1e-6) * self.k_weight[:, None]
 
         if cache is not None:
             q = self.rope(q, offset=cache.offset)
@@ -556,7 +564,6 @@ class PlamoModel(nn.Module):
 
 
 class Model(nn.Module):
-
     def __init__(self, config: ModelArgs) -> None:
         super().__init__()
         self.config = config
@@ -567,7 +574,7 @@ class Model(nn.Module):
 
         if not config.tie_word_embeddings:
             self.lm_head: nn.Module = nn.Linear(
-                config.hidden_size, vocab_size, bias=False
+                config.hidden_size, self.vocab_size, bias=False
             )
 
     def sanitize(self, weights: dict[Any, Any]) -> dict[Any, Any]:

llms/mlx_lm/models/su_rope.py

@@ -51,11 +51,13 @@ class SuScaledRotaryEmbedding(nn.Module):
             + math.log(max_position_embeddings / original_max_position_embeddings)
             / math.log(original_max_position_embeddings)
         )
+        self.dim = dims
 
     def __call__(self, x, offset: int = 0):
+        x[..., : self.dim] = self.scale * x[..., : self.dim]
         return mx.fast.rope(
-            self.scale * x,
+            x,
-            x.shape[-1],
+            self.dim,
             traditional=False,
             base=None,
             scale=1.0,

llms/mlx_lm/tuner/utils.py

@@ -98,6 +98,7 @@ def linear_to_lora_layers(
         "minicpm",
         "deepseek",
         "olmo2",
+        "olmoe",
         "internlm3",
     ]:
         keys = set(["self_attn.q_proj", "self_attn.v_proj"])
@@ -106,6 +107,8 @@ def linear_to_lora_layers(
         if model.model_type == "qwen2_moe":
             keys.add("mlp.gate")
             keys.add("mlp.shared_expert_gate")
+        if model.model_type == "olmoe":
+            keys.add("mlp.gate")
 
     elif model.model_type == "gpt_bigcode":
         keys = set(["attn.c_attn"])

llms/mlx_lm/utils.py

@@ -191,6 +191,7 @@ def get_model_path(path_or_hf_repo: str, revision: Optional[str] = None) -> Path
                         "*.py",
                         "tokenizer.model",
                         "*.tiktoken",
+                        "tiktoken.model",
                         "*.txt",
                         "*.jsonl",
                     ],
@@ -1015,6 +1016,46 @@ def save_config(
         json.dump(config, fid, indent=4)
 
 
+def mixed_quant_predicate_builder(
+    low_bits: int = 4, high_bits: int = 4, group_size: int = 64
+) -> Callable[[str, nn.Module, dict], Union[bool, dict]]:
+    def mixed_quant_predicate(
+        path: str,
+        module: nn.Module,
+        config: dict,
+    ) -> Union[bool, dict]:
+        """Implements mixed quantization predicates with similar choices to, for example, llama.cpp's Q4_K_M.
+        Ref: https://github.com/ggerganov/llama.cpp/blob/917786f43d0f29b7c77a0c56767c0fa4df68b1c5/src/llama.cpp#L5265
+        By Alex Barron: https://gist.github.com/barronalex/84addb8078be21969f1690c1454855f3
+        """
+
+        if not hasattr(module, "to_quantized"):
+            return False
+
+        index = int(path.split(".")[2]) if len(path.split(".")) > 2 else 0
+
+        num_layers = config["num_hidden_layers"]
+        use_more_bits = (
+            index < num_layers // 8
+            or index >= 7 * num_layers // 8
+            or (index - num_layers // 8) % 3 == 2
+        )
+        if "v_proj" in path and use_more_bits:
+            return {"group_size": group_size, "bits": high_bits}
+        if "down_proj" in path and use_more_bits:
+            return {"group_size": group_size, "bits": high_bits}
+        if "lm_head" in path:
+            return {"group_size": group_size, "bits": high_bits}
+
+        return {"group_size": group_size, "bits": low_bits}
+
+    return mixed_quant_predicate
+
+
+mixed_3_6 = mixed_quant_predicate_builder(low_bits=3)
+mixed_2_6 = mixed_quant_predicate_builder(low_bits=2)
+
+
 def convert(
     hf_path: str,
     mlx_path: str = "mlx_model",

llms/tests/test_prompt_cache.py

@@ -298,7 +298,7 @@ class TestPromptCache(unittest.TestCase):
         ):
             i += 1
             self.assertEqual(tok, toks[i])
-            self.assertTrue(mx.allclose(logits, all_logits[i], rtol=2e-2))
+            self.assertTrue(mx.allclose(logits, all_logits[i], rtol=3e-2))
 
 
 if __name__ == "__main__":

lora/lora.py

@@ -3,6 +3,7 @@
 import argparse
 import json
 import math
+import sys
 import time
 from pathlib import Path
 
@@ -14,6 +15,9 @@ import utils as lora_utils
 from mlx.utils import tree_flatten
 from models import LoRALinear
 
+# Disable output buffering to see print statements in real-time
+sys.stdout.reconfigure(line_buffering=True)
+
 
 def build_parser():
     parser = argparse.ArgumentParser(description="LoRA or QLoRA finetuning.")